Vehicular Carriers: When Data Meet the Road1

Prométhée Spathis, UPMC Marcelo Dias de Amorim, Raul A. Gorcitz, Yesid Jarma,

Serge Fdida, UPMC Ryuji Wakikawa,Toyota ITC

John Whitbeck, Vania Conan, Thales 1 Vehicular Carriers for Big Data Transfers, in Proc. of IEEE VNC2012, Nov 2012, Seoul Korea

Why Study Vehicules?

•  Present in large numbers –  Already surpassed 1 billion mark (2010) –  Expected to double in the next two decades

•  Mobile by nature, obviously –  Transportation of goods and people –  Public vehicles as urban sensors

•  Well instrumented and connected –  Sensors, communication devices and computing units –  Already exploited by manufactures to provide

advanced services

2

VANETs vs MANETs

3

MANET VANET

# of nodes 100s to 1000 Can be up to 1,000,000 vehicles

Area of movement 1,000,000 m2 Unbounded

(country wide)

Mobility Low to medium High

Trajectories Random waypoint One dimensional

Distribution Random and uniform

Sparse and uneven

VANETs’ New Apps

•  Safe navigation and autonomous driving: –  Vehicle & Vehicle, Vehicle & Roadway communications –  Forward Collision Warning, Blind Spot Warning, Intersection

Collision Warning

•  Entertainment –  Share location critical multimedia files –  Exchange local ad information, points of interest –  Support passenger to passenger internet games

•  Smart City Applications –  Monitor pollution and optimize traffic flow –  Smart Navigation Services –  Urban Surveillance

4

Research in VANETs

•  MAC & Physical Layer –  DSRC –  IEEE 802.11p

•  Data Dissemination –  More later

•  Mobility Models & Simulators –  Traffic Simulators –  Real Traces

•  Security & Privacy –  False Data –  DOS Attacks –  Privacy

5

The World of Big Data

6

Data Management

•  Large companies, organizations, universities, and governmental agencies

•  Technical flexibility and cost-effective scalability

•  Balance workloads, fast data retrieval, data replication, resources consolidation

•  Cloud computing, multimedia transfers, data migration, disaster recovery, online backups, file/email archiving

7 8

Vehicles as Data Carriers

Vehicular-based opportunistic bulk delay-tolerant data transfers using offloading technique

9

Autonomy Predictable itinerary

Room for storage capacity Mobile Modular Data Centers

10

Dataset

•  Annual Average Daily Traffic (AADT)

•  Provided by the French Ministry of Ecology, Energy, and Sustainable Development

•  Used for transportation planning and engineering

•  Measured from 2010 to 2011 •  Multiple point of

measurements per highways

11 12

Vehicle Density

4

one point on the highway, can be expressed in terms of thevehicular density as follows:

f = sf ⇥✓

k � k2

K

◆, (1)

where sf is the free-flow speed (i.e., maximum speed), K isthe density of vehicles characterizing a traffic jam, and k isthe actual density measured in vehicles per kilometer (a trafficjam takes place when the value of k approaches K). Therefore,the total transfer latency ⌧ can be expressed as a function ofthe vehicular density as follows:

⌧ =D ⇥ K

sf ⇥ ((k ⇥ K) � k2) ⇥ S ⇥ P +d

s̄· (2)

In Fig. 2, we present the transfer latency for transporting1 PB of data in function of the density of vehicles usingEquation 2. We consider vehicles with storage capacities of250 and 1,000 Gigabytes, with a class A level of service on arural highway with 3 lanes [12].2 On one extreme, long delaysare obtained with low traffic densities as less vehicles areavailable to act as carriers. As density increases, we observe asignificant increase in performance until we reach the optimaldensity. On the other extreme, as vehicle density grows beyondthe optimal value, congestion levels rise and the speed of thevehicle flow slowly decreases until it reaches a jam zone thatcauses a steep increase of the latency parameter.

In Fig. 2, the vertical line indicates the density obtainedfrom the dataset (5.96 vehicles/Km), which is the referencevalue we will use throughout the rest of the paper. Note that,in this case, even though the measured flow is largely belowthe optimal capacity of the highway, the transfer latency valuesobtained are as low as 8 hours to transfer 1 PB of data, and hasthe potential of being lower as the vehicle density approachesthe design optimum.

In Fig. 3(a), we show the transfer latency as a function of thetotal amount of data to be transferred. The two curves representvehicles equipped with 250-GB and 1-TB storage unities. Ourproposal is able to obtain delays of under 9 hours for quantitiesof data running up to 1 PB using a storage capacity of 1 TB.As we will see later in this paper, these values overcome byfar the performance obtained by alternative solutions that relyon the current Internet architecture. In Fig. 3(b), we show thethroughput of the system in function of the total data to betransferred. We also derive the theoretical throughput of thesystem and show the results in Fig. 3(b).

Other highways. We have performed the same calculations fora number of highways in France and show the results Fig. 4.As we can see, the values shown in the figure are compliantwith the numbers obtained for Orleans$Tours.

2The “class A” level of service refers to very good driving conditions wherethe drivers are unhindered by other traffic participants and are able to maintainthe desired speed. Vehicle density values are the main performance metric usedfor the level of service estimation.

0

5

10

15

20

25

30

400000 700000 1e+06

Late

ncy

(hou

rs)

Total data (GB)

250GB1TB

(a) Transfer latency

60

80

100

120

140

160

180

200

220

240

260

280

400000 700000 1e+06

Syste

m thr

ough

put (

Gbps

)

Total data (GB)

250GB1TB

(b) Throughput

Fig. 3. Transfer latency and throughput of the system for the road connectingOrleans and Tours. We consider different values of D and S. Note that thecurves are not linear because of the variable term in Equation 2.

Lille

Paris

Le Mans

AngersDijon

Lyon

Valence

Avignon

Tours

⌧ =A = 32, 935

9.2 h

46,2906.4 h⌧ =

A =

6.6 h52,390

⌧ =A =

12.2 h21,408

⌧ =A =

8.2 h33,922

⌧ =A =

10.5 h31,753

⌧ =A =

6.4 h53,196

⌧ =A =

4.7 h64,535

⌧ =A =

4.7 h67,813

⌧ =A =

Orleans

Fig. 4. Average Transfer Delays (⌧ ) obtained using the “annual average dailytraffic” (A) on segments of highways connecting several important locationsin France. The parameters used here are: average speed = 100 Km/h, totaldata = 1 PB, per-vehicle storage capacity = 1 TB, and penetration ratio =20%.

C. Impact of “environmental parameters”

As we have seen from Equation 2, other than the vehicledensity and the amount of data to be transferred, the perfor-mance of our system depends as well on other “environmental”factors such as distance and speed. Therefore, it is necessaryto assess the impact of both these factors.

1) Average speed s̄: In order to better understand theimpact of the average vehicle speed, we have considered thesame parameters as before with the exception of the total datato be transferred, which has been fixed at 1 PB, and the datastorage capacity which has also been fixed at 1 TB. We varythe average speed from 60 Km/h to 130 Km/h. The results areshown in Fig. 5. It is clear that, even if the average speed hasa non-negligible impact on the performance of the solution,the system is relatively impervious to changes in the average

Computing Transfer Latency

13

Transfer Latency

Total Data

Penetration Ratio

Storage Capacity

Distance

Avg Speed

Highway Capacity vs Vehicle Density

14

0

40

80

120

160

0 10 20 30 40 50 60 70

Late

ncy

(hou

rs)

Vehicle density (Vehicles/Km)

Measured value (5.96 Vehicles/Km)

250GB1TB

1 Pbytes data – Class A vehicles – 3-lane rural highway

Intercity scenario: Tours – Orléans

15

Vehicle Flow = 706 Veh/H

Penetration Rate = 20%

Highway Length = 118 Km

Average Speed = 100 Km/H

0

5

10

15

20

25

30

400000 700000 1e+06

Late

ncy

(hou

rs)

Total data (GB)

250GB1TB

60

80

100

120

140

160

180

200

220

240

260

280

400000 700000 1e+06

Syst

em th

roug

hput

(Gbp

s)

Total data (GB)

250GB1TB

Transfer latency for delivering up to 1 PB of data between Orleans and Tours using vehicle carriers.!

System throughput for delivering up to 1 PB of data using the delay constraints obtained in Figure 1.!

16

Average Transfers Delay Lille

Paris

Le Mans

AngersDijon

Lyon

Valence

Avignon

Tours

⌧ =A = 32, 935

9.2 h

46,2906.4 h⌧ =

A =

6.6 h52,390

⌧ =A =

12.2 h21,408

⌧ =A =

8.2 h33,922

⌧ =A =

10.5 h31,753

⌧ =A =

6.4 h53,196

⌧ =A =

4.7 h64,535

⌧ =A =

4.7 h67,813

⌧ =A =

Orleans

Α "Annual Average Daily Traffic

τ Average Transfer Latency

Impact of Speed and Distance

17

0

3

6

9

60 70 80 90 100 110 120 130

Late

ncy

(hou

rs)

Average vehicle speed (Km/h)

21.6% 19.1% 16.3% 15.6% 14.2% 13.1% 12.1% 11.2%

1TBSpeed influence proportion

0

3

6

9

12

50 100 150 200 250 300 350 400

Late

ncy

(hou

rs)

Distance on highway (Km)

6.5% 11.3% 17.4% 22% 26% 29.7% 33% 36.1%

1TBDistance influence proportion

Cost Comparison

18

Cost comparison between a full electrical recharge cost and a package delivery system.!

Cost comparison between Internet with dedicated links and Vehicular Carriers.

600

800

1000

1200

1400

1600

1800

2000

400000 700000 1e+06

Cos

t (in

Eur

os)

Total data (GB)

Energy costUPS cost

0

10000

20000

30000

40000

50000

60000

400000 700000 1e+06

Cos

t (in

Eur

os)

Total data (GB)

Internet (1TB) Internet (250GB)

Energy (1TB) Energy (250GB)

Conclusion

•  Offloading delay tolerant data to vehicles could ease the burden on the legacy Internet infrastructure

•  Vehicular Carriers can be higly efficient in terms of transfer latency when large amounts of data are considered

•  Incentive based motivational actions can increase the penetration ratio of the system

19

Open Issues

•  Routing •  Delay sensitiveness •  Context •  Resilience/robustness •  Reliability •  Security •  Incentives •  Business model

20